Re: glibc changes

[Marcus Brinkmann]

At Wed, 9 Feb 2005 13:03:17 -0500,
Barry deFreese wrote:
> I'm looking to try to do some updates to glibc.  This started as the missing 
> weak_aliases but if possible may grow into the sysv and POSIX semaphore 
> stuff also.
> 
> Roland, marcus, etc, can someone comment on the following patch neal made 
> two years ago for sysv shared memory?
>
> http://lists.gnu.org/archive/html/bug-hurd/2002-04/msg00288.html

On retrospect, I wonder if the cleanest approach to the "extra
statistics" problem wouldn't be to just define some .gnu.shm.* xattr's
that provide them, and implement RPCs to retrieve xattr's from
filesystem servers.  We can fallback to some dummy statistics for
filesystems that don't implement them, so that you can still use dumb
filesystems (without special shm support) for shm, and you can use a
proper shmfs translator which offers the additional statistics as
xattrs if you want better compatibility.

This would require somebody to come up with an xattr RPC interface,
but that is generally useful, not just for SHM.

One idea for an xattr interface would be to just use dir_lookup
(""...) with some new open flag O_XATTR that opens a "hidden
directory" that contains all the xattrs, and you can list them like
any directory and read/write them as files.

Another way would be to add RPCs like get/set translator, just for
xattr.  Then we would need all operations: get, set, list, remove.

Roland, what do you think?  New operations would be more convenient to
use, but somehow duplicate the filesystem interface.  In addition,
when limits on xattr's are lifted, and they can be large files
themselves, a real filesystem interface for manipulating them would
probably come in handy.

Thanks,
Marcus